A central pillar of the Belmont Report is that a bright line must be drawn between medical practice and biomedical research. That line may have been brighter 50 years ago. Today, the typical physician is likely to work for a corporation or health system that styles itself as a learning health system. Such systems increasingly emphasize the (research-like) use of data to measure quality, encourage efficiency, ensure safety, and guide a standardized approach to clinical care. While these activities are not considered research, they pose many of the same risks or conflicts of loyalty. In research, the doctor's fiduciary loyalty to the patient is compromised by a loyalty to the scientific process. In learning health systems, the doctor's loyalty is compromised by loyalty to the system and its metrics. In this world, it is not clear that research-as conceptualized by the Belmont Report, codified in the Common Rule, and overseen by IRBs-is a uniquely risky activity deserving of such uniquely strict oversight. Perhaps, instead, the divided loyalties and conflicts of interest faced by everyday clinicians working in learning health systems demand a protective framework similar to the one that we now have for the activities that we designate as "research." This article compares the risks of the various activities that might be called "research" and suggests a unified system of oversight for all of them.